Clinically relevant medical concept clustering

• US 10,839,947 B2
• Filed: 01/06/2016
• Issued: 11/17/2020
• Est. Priority Date: 01/06/2016
• Status: Active Grant

• Current Assignee: Merative US LP
• Original Assignee: International Business Machines Corporation
• Inventors: Barker, Kenneth J., Devarakonda, Murthy V., Tsou, Ching-Huei
• Primary Examiner(s): Weisenfeld, Aryan E
• Assistant Examiner(s): Blanchette, Joshua B
• Application Number:
  US14/988,787
• Publication Number:
  US 20170193185A1
• Time in Patent Office: 1,777 Days
• Field of Search: 705 3
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/334 Query execution G06F16/335 takes precedence
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

The present invention embodiments are directed to methods, systems, and computer programs for identifying relations, within at least one taxonomy, between taxonomy categories and concepts extracted from electronic content. The relations represent semantic similarities for the concepts. The concepts are clustered based on the identified relations within the at least one taxonomy.

Claims

1. A computer program product for clustering concepts extracted from electronic content, the computer program product comprising one or more non-transitory computer readable storage media collectively having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

identify, within a plurality of different taxonomies, relations between taxonomy categories of the different taxonomies and the concepts extracted from the electronic content, wherein the electronic content is from a medical record, the concepts include medical concepts extracted from the medical record, and the plurality of different taxonomies includes medical taxonomies, and wherein the relations represent semantic similarities for the concepts and the identifying relations further includes;

mapping the concepts to each of the plurality of different taxonomies, wherein mapping the concepts includes;

determining a first concept extracted from the electronic content not found in a selected taxonomy of the plurality of different taxonomies;

identifying one or more other taxonomies of the plurality of different taxonomies containing the first concept and determining a second concept that resides in the selected taxonomy and the identified one or more other taxonomies; and

mapping the first concept to the second concept within the selected taxonomy when the second concept is closest to the first concept in the identified one or more other taxonomies and within a distance limit of the first concept, wherein the first concept remains unmapped to the selected taxonomy in response to the second concept not satisfying the distance limit;

generating concept vectors relating each of the concepts to one or more corresponding taxonomy categories of the different taxonomies, wherein each concept vector is associated with a concept and includes a plurality of values with each value indicating a relationship between that associated concept and a corresponding taxonomy category, and wherein at least one concept has relations to taxonomy categories in two or more taxonomies; and

determining a similarity measure between each of the concept vectors of the concepts based on distances between the concept vectors;

cluster the concepts based on the determined similarity measure between the concept vectors; and

generate a visualization of the electronic content with information arranged according to the clustered concepts to identify information within the electronic content relevant to a situation.

2. The computer program product of claim 1, wherein the program instructions further cause the processor to:

perform named-entity recognition and disambiguation on the concepts, which includes concept identification, named-entity detection, and linking of each identified concept to a meaning.

3. The computer program product of claim 1, wherein the program instructions further cause the processor to:

identify the concepts from both structured information and unstructured information within the electronic content.

4. The computer program product of claim 1, wherein the taxonomy categories represent a feature space for clustering of the concepts, and wherein the program instructions further cause the processor to:

perform dimensionality reduction to remove features from the feature space to reduce processing time.

11. The computer program product of claim 1, wherein the program instructions further cause the processor to:

generate a similarity matrix relating the concepts based on the similarity measures; and

cluster the concepts based on the similarity measures.

12. The computer program product of claim 1, wherein the semantic similarities for the concepts represent relative relationships of the concepts to the taxonomy categories such that the concepts are clustered based on identified relevance.

5. A system comprising:

at least one processor configured to;

identify, within a plurality of different taxonomies, relations between taxonomy categories of the different taxonomies and concepts extracted from electronic content, wherein the electronic content is from a medical record, the concepts include medical concepts extracted from the medical record, and the plurality of different taxonomies includes medical taxonomies, and wherein the relations represent semantic similarities for the concepts and the identifying relations further includes;

mapping the concepts to each of the plurality of different taxonomies, wherein mapping the concepts includes;

determining a first concept extracted from the electronic content not found in a selected taxonomy of the plurality of different taxonomies;

identifying one or more other taxonomies of the plurality of different taxonomies containing the first concept and determining a second concept that resides in the selected taxonomy and the identified one or more other taxonomies; and

mapping the first concept to the second concept within the selected taxonomy when the second concept is closest to the first concept in the identified one or more other taxonomies and within a distance limit of the first concept, wherein the first concept remains unmapped to the selected taxonomy in response to the second concept not satisfying the distance limit;

generating concept vectors relating each of the concepts to one or more corresponding taxonomy categories of the different taxonomies, wherein each concept vector is associated with a concept and includes a plurality of values with each value indicating a relationship between that associated concept and a corresponding taxonomy category, and wherein at least one concept has relations to taxonomy categories in two or more taxonomies; and

determining a similarity measure between each of the concept vectors of the concepts based on distances between the concept vectors;

cluster the concepts based on the determined similarity measure between the concept vectors; and

generate a visualization of the electronic content with information arranged according to the clustered concepts to identify information within the electronic content relevant to a situation.

6. The system of claim 5, wherein the at least one processor is further configured to:

identify the concepts from both structured information and unstructured information within the electronic content.

7. The system of claim 5, wherein the at least one processor is further configured to:

perform named-entity recognition and disambiguation on the concepts, which includes concept identification, named-entity detection, and linking of each identified concept to a meaning.

8. The system of claim 5, wherein the taxonomy categories represent a feature space for clustering of the concepts, and wherein the at least one processor is further configured to:

perform dimensionality reduction to remove features from the feature space to reduce processing time.

9. The system of claim 5, wherein the at least one processor is further configured to:

generate a similarity matrix relating the concepts based on the similarity measures; and

cluster the concepts based on the similarity measures.

10. The system of claim 5, wherein the semantic similarities for the concepts represent relative relationships of the concepts to the taxonomy categories such that the concepts are clustered based on identified relevance.